Adding Up Both Full Paths

Ada re-derives this chapter’s own numbers step by step, at full precision

ADA · CALCULATION AUDIT

Adding Up Both Full Paths

The chapter’s trace lists a 12 ms cloud kernel, an 80-120 ms WAN path, an 18 ms cloudlet kernel, and a 4 ms local hop — then says the slower kernel still wins. The kernel-only view favours the cloud by 6 ms, but the full loop favours the cloudlet by 70 to 110 ms. This audit totals both full paths to show why the acceptance gate belongs on the end-to-end path, not the datasheet kernel time.

Companion to the chapter Cloudlets: Local Compute Pools — every number here comes from that chapter.

Ada: The trace above lists a 12 ms cloud kernel, an 80-120 ms WAN path, an 18 ms cloudlet kernel, and a 4 ms local hop, then says the slower kernel still wins. That is exactly the kind of claim that looks wrong until you add the terms, so let me total both interaction paths.

• Cloud full path. 12 ms kernel + 80 ms network = 92 ms at best; 12 + 120 = 132 ms at worst.
• Cloudlet full path. 18 ms kernel + 4 ms network = 22 ms.
• Compare. The kernel-only view favours the cloud by 18 - 12 = 6 ms. The full loop favours the cloudlet by 92 - 22 = 70 ms (best case) to 132 - 22 = 110 ms (worst case) — a 92 / 22 = 4.2x to 132 / 22 = 6.0x speedup.

The 6 ms kernel advantage is both real and irrelevant: the network term is an order of magnitude larger and arrives before every result. That is why the acceptance gate belongs on the measured end-to-end path, not the accelerator’s datasheet kernel time.

Every number above is taken from the chapter’s own material and re-derived step by step.